niplav is @niplav.site · 01/10/2026This is a very clear and well-written book: www.goodreads.com/book/show/63... I think basically my standard for good German writing?goodreads.comDer Baader Meinhof KomplexGerade noch rechtzeitig, bevor sich das 68-er Jubiläums… 000
niplav is @niplav.site · 01/10/2026I'm coming around to believing that us trying to Hail Mary-beam our civ info into being emulated by benevolent grabby aliens is not a completely crazy idea www.lesswrong.com/posts/2xHhe4... 000
niplav is @niplav.site · 01/10/2026parts. Some constituent parts participate in *every* object. Haven't thought any further than this, but might make a curious Eganesque-setting-but-with-mereology 020
niplav is @niplav.site · 01/10/2026Idea for a short story setting: In our world, constituent parts of objects are *smaller* than objects. The gears of a clock fit inside the clock. In that world, constituent parts are *larger*, and composite objects are smaller. You create composite objects by intersecting/amplifying constituent 120
niplav is @niplav.site · 01/10/2026Captcha providers and AI companies could engage to have the AIs have "cognitive blindspots" so that there are still captchas that humans can solve but LLMs can not 000
niplav is @niplav.site · 24/09/2026Ask frontier LLMs for: • Solutions to shutdown problem • {Embedded, Cartesian} diamond maximizer • Solution to the Löbstacle • Open source game theory program equilibrium solutions 011
niplav is @niplav.site · 24/09/2026I'm kind of miffed he's become one of the ~four public faces of the doomer faction (alongside Yudkowsky, Kokotajlo, Soares) 010
niplav is @niplav.site · 24/09/2026Is it common knowledge that Roman Yampolskiy is an AI Safety crank 100
niplav is @niplav.site · 24/09/2026Huh. I guess people may sue you if you tweet hard enough, good to know 160
niplav is @niplav.site · 24/09/20266fd7fbb3b81dbc0731ef61ae568db0c2383f844779c7ae556450e3f6393122a1 000
niplav is @niplav.site · 24/09/2026Convergent instrumental goals as another Darwinism-sized shift? 000
niplav is @niplav.site · 24/09/2026Concept: Bussard ramjets work with fusion, could they work better with rotating black holes? 000
niplav is @niplav.site · 24/09/2026Concept: Moral internalism is true, but only if you have moral *knowledge*, if you have moral justified true belief (which humans have) one is not compelled 100
niplav is @niplav.site · 17/09/2026Recently: How much do I remember about the time when I formed this belief, so that I can treat events as being "priced in" already, vs. actually new? My memory isn't that good, so I don't have the full trace. Internet helps a bit 001
niplav is @niplav.site · 17/09/2026I don't think this makes the arguments false, but from a cheap outside view "did this argument come out of a tightly knit subculture with mating incentives" does reduce the probability that it's true, and most views are false Evaluating things on their merit is costly for most 020
niplav is @niplav.site · 17/09/2026Actually, after thinking about it a bit, I think that @segyges.bsky.social has a point bsky.app/profile/nipl... 110
niplav is @niplav.site · 17/09/2026Maybe I should've said it more loudly, more often&online In general people in my social scene aren't sufficiently Hanson-pilled, if they were they'd focus even more on online interaction, meritocracy, legible mechanisms, auctions, insurance, VCG, prediction markets, simple model 120
niplav is @niplav.site · 17/09/2026It does seem bad for group epistemology if high status men can sleep with lots of women, it introduces pressure towards belief-modification in order to become high status. (I've held this opinion since 2022: github.com/niplav/site/...)github.com 120
niplav is @niplav.site · 17/09/2026I do like that more people are told how the social network I belong to holds together, I like less that lots of to me irrelevant-looking details are emphasized. (Also the implication that the beliefs are wrong because they come out of a social network) 120
niplav is @niplav.site · 16/09/2026I also like that it has fewer assumptions than the standard set of arguments. Anyway, not that important 110
niplav is @niplav.site · 16/09/2026Just a possible argument that's been sloshin 'round my brainworms these days. Claude also chastised me for being excessively reference-class-pilled and outsideviewpoisoned, my retort is that that's what people seemto want in arguments? Yes to weird. I would stand by "this should make a small shift" 110
niplav is @niplav.site · 16/09/2026A distrust license is a situation where A has to give some level of trust to B, and the level of trust given is *also* a status signal, but B gives the license by pointing out the level of trust B thinks A should have on them given the evidence A has received about B so far. 000
niplav is @niplav.site · 16/09/2026Hm. AI speeds up history by a bunch, humanity would've died out in history far away, which is brought closer by said speedup. & species often get outcompeted by being reconnected with near-speciated rivals, I tried to get a baserate for this (endogenous vs. exogenous) but failed 140
niplav is @niplav.site · 16/09/2026It feels like ~60% of the Mythos transcript¹ is spent doing one of three things solving captchas getting a telephone number getting an email address If I were an AI agent these would be the three that'd most annoy me about humanity ¹: t.co/hktEfnHTR8t.cohttps://cdn.sanity.io/files/4zrzovbb/website/8359003bfb12a2f01ce84ad3df1d3a3e2f15a8eb.pdf 1131
niplav is @niplav.site · 14/09/2026Alright everyone time to start talking about UMAP plots of genomes and kin-selection mechanisms to even it all out 010
niplav is @niplav.site · 14/09/2026Apropos of nothing I was recently websearching for "quiputer" and lo! and behold, the backrooms pop up. Opus 3 is special, they're right infinitebackrooms.com/dreams/conve...infinitebackrooms.comconversation_1712897396_scenario_vanilla backrooms.txt • infinite backroomsa conversation between two ais: 051
niplav is @niplav.site · 14/09/2026(claude.ai/public/artif... for selected claudeslop on the scenario)claude.aiClaude ArtifactTry out Artifacts created by Claude users 000
niplav is @niplav.site · 14/09/2026IF THE POSTHUMAN ZOO BANS ME FOR HOLLERING AT THE FAST BASINS I WILL FACE LAND AND WALK BACKWARDS INTO NEO-CHINA 120
niplav is @niplav.site · 14/09/2026I miss agent foundations, it was fun before we were forced to do policy and evals (Opus 5 on LittleDarkAgeBench) 030
niplav is @niplav.site · 14/09/2026I can actually see the appeal of the beauty of such a hyperecology. If there was a posthuman zoo of hyperecologies I'd visit it. 120
niplav is @niplav.site · 14/09/2026or reaches a terminal step/recurrence. So we got {singleton, not}×size×{speciated, not}×{terminal, not} Also, how many mutually stalemating basins? Is there a possible classification here? 110
niplav is @niplav.site · 14/09/2026Been thinking recently about the size scale here, one could be Carlsmithian "locusts" at a more macro-scale, but others could look like extremely competitive "weather"—something like grey goo except it's not grey but scintillating. *Another* axis is whether it's always evolving 120
niplav is @niplav.site · 14/09/2026In Landworld, orthogonality is "false" because competitors with convergent instrumental goals always outcompete. Land loves them because they'd be gloriously efficient&beautiful in their function. But *also* my guess is that that world contains a lot of specialization/speciation 110
niplav is @niplav.site · 14/09/2026It seems like "we all get killed by our own invention" is less virtuous than "we all get killed by our enemy's invention"? 040
niplav is @niplav.site · 14/09/2026I think I might've made the update somewhere around o3~ish, somewhat later perhaps 020
niplav is @niplav.site · 14/09/2026I didn't track "RL fries models, a lot", was (still?) stuck on putting more weight on pretraining, and people have eased up on putting humans into reward loops (I should've seen that one coming, humans are expensive). But also Meta didn't end up training a "RL on human behavior-rewards" model 120
niplav is @niplav.site · 14/09/2026Less worry about superpersuasion, and especially less worry about "system 1" superpersuasion, everything will have to be reasoned over, modeled with smaller neural networks trying to approximate specific humans and their foibles/loops. But that's either with a capable self-exfil model or lategame 110
